前言
下文介绍的自定义协议仅作为学习示例,纯粹是玩具项目,没有实际可用性。无需过度关注和讨论其合理性
进行通信的双方是谁?
常见的模型
客户端-服务器,例如HTTP协议,浏览器<=>Web服务器。
中转站模型,如MQTT协议,应用服务<=>中转站<=>硬件客户端
对等模型,例如Thrift协议,应用服务<=>应用服务。
通用协议如此丰富,还需要自定义协议吗?
需要。许多中间件服务在构建集群时,服务节点之间需要进行高效的内部通信。
在这种场景下,自定义协议能发挥巨大的作用:
- 去除冗余字段:自定义协议能够减少无用字段,最大化优化通信吞吐量
- 灵活性:自定义协议可以根据需求进行灵活扩展,支持注入优先级控制,解压缩控制等特点。
自定义协议可以减少无用字段,最大限度地优化通信吞吐量;也更加灵活,可以进行优先级控制。
例如,Kafka 就使用了自定义协议来满足高效的消息传递需求。
自定义协议设计
所谓网络协议,就是传输的报文格式,以及收发双方处理报文的规则。
报文格式做如下设计:
- 固定头部(4字节)
- 字节1:消息类型
- 1=req,2=res, 3=pub, 4=sub, 5=msg
- 用一个字节来表示类型有点浪费了。
- 字节2~字节4:消息体长度
- 这三个字节能够表示最大值为 16777215,即最大消息体长度为 16MB。
- 字节1:消息类型
- 消息体(可变长度)
规则:
1.服务端收到req包,需返回res包
2.服务端收到sub包,需更新订阅情况
3.服务端收到pub包,需根据订阅情况发送msg包
粘拆包问题
在设计网络协议时,不可不谈粘拆包问题。
什么是粘包和拆包?
这两个都是接收端在接收数据时遇到的问题,其中
- 粘包:多个数据包合并成一个包接收
- 拆包:一个数据包被拆分成多个包接收
为什么会出现粘包与拆包?
根本原因就是传输层的TCP协议,是面向字节流的,它不知道数据边界。
此外,TCP根据网络情况(如最大传输单元MTU)动态调整报文大小,导致数据包的分段与合并。
从而产生粘包和拆包问题
传输流程:
1.发送缓冲区:当应用层产生数据后,这些数据会首先进入Socket连接的发送缓冲区
2.数据拆分:网卡根据缓冲区中的数据内容,将数据拆分成多个小的TCP数据报进行发送
3.接收与重组:接收端的TCP栈会将接收到的多个TCP数据包重新组装成完整的字节流(Socket接收缓冲区)
案例场景
一个常见的场景是,客户端连续发送多个消息(如 100 个字符串),而服务端接收到的数据可能并不完全是 100 条。
要复现这种问题也很简单,只要客户端连续发100个字符串,检查服务端收到的数据条数。
客户端代码:连接建立后,连续发送100次字符串
@Override
public void channelActive(ChannelHandlerContext ctx) throws Exception {
// ctx.writeAndFlush(Unpooled.copiedBuffer("Netty rocks!", CharsetUtil.UTF_8));
new Thread(() -> {
for (int i = 0; i < 100; i++) {
ctx.writeAndFlush(Unpooled.copiedBuffer("Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!", CharsetUtil.UTF_8));
}
}).start();
}
服务端代码:每收到一个包,就打印一次。
@Override
public void channelRead(ChannelHandlerContext ctx, Object msg) throws Exception {
ByteBuf in = (ByteBuf) msg;
System.out.println("Server Receive:"+in.toString(CharsetUtil.UTF_8));
ctx.write(in);
}
结果:仅收到两个包,同时存在粘包和拆包问题。一个Siuuuu被截断了
Server Receive:Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuu
Server Receive:uuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!Siuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu!
如何处理粘包和拆包?
处理方式由消息格式决定
- 固定长度:每条消息的长度固定,不足部分使用填充
- 特殊分隔符:每条消息的末尾添加特定的分隔符
- 消息头+消息体:消息头长度固定,包含消息体长度信息
由于我们采用的时第三种方式,也是最复杂的一种。
处理的核心在于消息头,因为它携带了消息体的长度信息,是判断消息边界的关键。
粘包的处理
步骤如下:
- 提取消息头:首先提取消息头,从中获取消息体的长度信息
- 读取完整消息:根据消息体的长度,从数据流中读取完整的消息内容
- 重复执行:重复步骤1和步骤2,直到没有更多的数据,或当前数据不足以构成完整的消息
拆包的处理
拆包的处理方式与粘包类似:
- 缓存数据:如果接收到的数据不足一条完整消息,则将数据存入缓冲区。
- 合并新数据:在接收到新数据时,判断缓冲区和新数据是否可以组成完整消息,直到消息完整为止。
- 继续缓存:剩下的数据如果不足,则继续缓存
代码案例
1)客户端
根据上面的协议格式,构建消息。(这里的消息体内容是随机字符串,实际应用中通常是序列化后的POJO对象。)
连接建立后连续发送200条随机长度的消息。
public class EchoClientHandler extends SimpleChannelInboundHandler<ByteBuf> {
@Override
public void channelInactive(ChannelHandlerContext ctx) throws Exception {
super.channelInactive(ctx);
System.out.println("断开连接");
}
@Override
public void channelActive(ChannelHandlerContext ctx) throws Exception {
new Thread(() -> {
//连续发送200条消息
for (int i = 0; i < 200; i++) {
try {
ctx.writeAndFlush(Unpooled.copiedBuffer(buildRandomMsg()));
} catch (IOException e) {
e.printStackTrace();
}
}
}).start();
}
protected void channelRead0(ChannelHandlerContext channelHandlerContext, ByteBuf byteBuf) throws Exception {
System.out.println("Client receive:"+byteBuf.toString(CharsetUtil.UTF_8));
}
@Override
public void exceptionCaught(ChannelHandlerContext ctx, Throwable cause) throws Exception {
cause.printStackTrace();
ctx.close();
}
//构建消息,其中body内容为随机长度的随机字符串
public static byte[] buildRandomMsg() throws IOException {
int length = RandomUtil.randomInt(100, 200);
String body = RandomUtil.randomString(length);
System.out.println("长度:"+length+"||内容:"+body);
byte type = 1;
byte[] lengthBytes = new byte[3];
lengthBytes[0] = (byte) (length >> 16);
lengthBytes[1] = (byte) (length >> 8);
lengthBytes[2] = (byte) length;
byte[] bodyBytes = body.getBytes(CharsetUtil.UTF_8);
return concatByteArrays(new byte[]{type}, lengthBytes, bodyBytes);
}
//拼接字节数组
public static byte[] concatByteArrays(byte[]... byteArrays) throws IOException {
// 使用 ByteArrayOutputStream 来拼接字节数组
ByteArrayOutputStream byteArrayOutputStream = new ByteArrayOutputStream();
for (byte[] array : byteArrays) {
byteArrayOutputStream.write(array);
}
// 返回拼接后的字节数组
return byteArrayOutputStream.toByteArray();
}
}
2)服务端
在看代码前,先说明一下channelRead的调用流程
- Socket接收到TCP报文,将数据写入内核缓冲区
- NIO线程检测到此Socket有可读消息
- NIO线程从内核缓冲区读取消息,得到ByteBuf msg
- NIO线程调用channelRead
得到两个信息
- msg是从缓冲区读取的,它可能包含多条完整消息 + 一条残缺消息。
- msg已经从缓冲区读出,缓冲区数据已清空。对于不完整的消息需要自行缓存
下面代码是直接实现的,主要用来介绍完整的处理逻辑。
实际应用中推荐继承Netty提供的ByteToMessageDecoder,它帮你实现了缓存管理。
public class EchoServerHandler extends ChannelInboundHandlerAdapter {
private static final int HEADER_LENGTH = 4; //消息头部长度
private ByteBuf buffer = Unpooled.buffer(1024); //缓存残缺消息
@Override
public void channelRead(ChannelHandlerContext ctx, Object msg) throws Exception {
ByteBuf income = (ByteBuf) msg;
//上一次有缓存存在,则本数据包不是消息头开头,
if(buffer.readableBytes() > 0) {
buffer.ensureWritable(income.readableBytes()); //进行必要的扩容
income.readBytes(buffer, income.readableBytes());
readMsgFromBuffer(buffer);
//剩下一点残缺消息
if(buffer.readableBytes() > 0) {
//保留剩下的数据,重置读索引为0
System.out.println("缓存剩余字节:"+buffer.readableBytes());
buffer.discardReadBytes();
} else { //刚刚好,则清空数据
buffer.clear();
}
} else {
readMsgFromBuffer(income);
//剩下的数据全部写入缓存
if (income.readableBytes() >0) {
System.out.println("剩余字节:"+income.readableBytes());
income.readBytes(buffer, income.readableBytes());
}
}
}
//从字节数组中读取完整的消息
private void readMsgFromBuffer(ByteBuf byteBuf) {
//剩余可读消息是否包含一个消息头
while(byteBuf.readableBytes() >= HEADER_LENGTH) {
byteBuf.markReaderIndex(); //由于可能读不到完整的消息,所以读之前先标记索引位置,方便重置
//读取消息头
byte[] headerBytes = new byte[4];
byteBuf.readBytes(headerBytes);
//获取类型
int type = headerBytes[0] & 0xFF;
//获取消息体长度
int bodyLength = ((headerBytes[1] & 0xFF) << 16) |
((headerBytes[2] & 0xFF) << 8) |
(headerBytes[3] & 0xFF);
//不包含请求体
if (byteBuf.readableBytes() < bodyLength) {
byteBuf.resetReaderIndex(); //重置读索引到当前消息头位置
break;
}
// 完整消息体已经接收,处理消息
byte[] body = new byte[bodyLength];
byteBuf.readBytes(body);
System.out.println("type:"+type+"||length:"+bodyLength+"||body:"+new String(body, CharsetUtil.UTF_8));
}
}
@Override
public void channelReadComplete(ChannelHandlerContext ctx) throws Exception {
// ctx.writeAndFlush(Unpooled.EMPTY_BUFFER).addListener(ChannelFutureListener.CLOSE);
ctx.writeAndFlush(Unpooled.EMPTY_BUFFER);
}
@Override
public void exceptionCaught(ChannelHandlerContext ctx, Throwable cause) throws Exception {
cause.printStackTrace();
ctx.close();
}
}
服务端输出:服务端逐行打印出消息类型,长度,消息体。
...
type:1||length:175||body:0cDDAkum0F9DNwF511AKitTe2zRoSc27IjBYwgoODkXxx78xp0cowcDDNWTZ6xjCZyn6wmI2UxXLYB25TjUnOG9ZyjiZ9Jge3kbxabRjZAo0qsCYFfKMyzxApp953z1N7uDbP9rmlxeyYbYiif3y3ybtnnaAkuKFcspje6SLRnY69Nz
消息体编解码(序列化)
在经过前面粘包和拆包处理后,我们已经能够成功地从数据流中分离并组装出完整的消息。然而,在实际应用中,消息体通常需要进一步转换为对象,才能提交给上层的业务逻辑。
这是传输层的关键职责之一。
常见序列化方法
常见的POJO对象序列化方式包括:
Java序列化(Serializable)
优点:内置,无需额外依赖。
缺点:
- 性能较差,序列化和反序列化速度较慢。
- 无法跨语言使用,限制了不同语言(如Java服务端和C++客户端)之间的数据交换。
JSON
优点:可读性好,方便调试,支持各种语言
缺点:相较于二进制格式,JSON的键(key)通常占用较多空间,大规模数据传输时,带宽开销大。
Protocol Buffers(ProtoBuf)
优势:
- 高效的二进制序列化,体积小,序列化和反序列化速度快。
- 支持跨语言使用,适用于不同编程语言之间的通信。
代码案例
这里我们使用ProtoBuf。
构建消息类
写一个.proto文件,定义消息格式。
hello_request.proto
option java_multiple_files = true;
option java_package = "protocol";
option java_outer_classname = "Request";
message HelloRequest {
required string requestId = 1;
optional string content = 2;
}
下载ProtoBuf编译工具包,protoc-{version}-win64.zip
https://github.com/protocolbuffers/protobuf/releases
编译,得到Java文件
protoc -I=$SRC_DIR --java_out=$DST_DIR $SRC_DIR/hello_request.proto
引入对应版本的Jar包。(jar包版本要和protoc版本一致,否则报错)
https://mvnrepository.com/artifact/com.google.protobuf/protobuf-java
接着就可以使用类构建POJO对象和对象的编解码了。
客户端
其他地方不变,使用上面生成好的HelloRequest类,构建对象。通过setter塞入数据,然后通过toByteArray()得到序列化后的二进制数据。
注意:现在的length应该是整个消息体的字节数,不再是随机字符串的长度。
public static byte[] buildRandomMsg() throws IOException {
int randomStrLength = RandomUtil.randomInt(100, 200);
String msgId = UUID.randomUUID().toString();
String content = RandomUtil.randomString(randomStrLength);
HelloRequest request = HelloRequest.newBuilder()
.setRequestId(msgId)
.setContent(content)
.build();
byte[] bodyBytes = request.toByteArray();
int length = bodyBytes.length;
System.out.println("发送消息:"+request.toString());
byte type = 1;
byte[] lengthBytes = new byte[3];
lengthBytes[0] = (byte) (length >> 16);
lengthBytes[1] = (byte) (length >> 8);
lengthBytes[2] = (byte) length;
return concatByteArrays(new byte[]{type}, lengthBytes, bodyBytes);
}
服务端
其他地方不变,解析body的时候,使用HelloRequest.parseFrom(byte[] bytes)进行解码,得到HellpRequest对象。
//System.out.println("type:"+type+"||length:"+bodyLength+"||body:"+new String(body, CharsetUtil.UTF_8));
if(type == 1) {
try {
HelloRequest request = HelloRequest.parseFrom(body);
System.out.println("收到消息:"+request.toString());
} catch (Exception e) {
System.out.println("解析失败:"+new String(body, CharsetUtil.UTF_8));
}
} else {
System.out.println("消息类型未知:"+type);
}
结果
客户端输出
...
发送消息:requestId: "ca9b3e07-0662-467c-9bed-843b519c2480"
content: "q82EuHvGgMhwbHl1t0qfv4M2NCJLikxahpEc8q9ezpCWUbU9M1Oh6U6zfIOnBC50ex5BweYfZ2JB0NoLmP4hgIsNzZ8mtfFPayi8KlDWRQw3gj7ENRgxjbm4HxJgrdDNobuguc8EPQ3SccWXGTsZytLEeOHJXskiGlH4oEf"
服务端输出
....
收到消息:requestId: "ca9b3e07-0662-467c-9bed-843b519c2480"
content: "q82EuHvGgMhwbHl1t0qfv4M2NCJLikxahpEc8q9ezpCWUbU9M1Oh6U6zfIOnBC50ex5BweYfZ2JB0NoLmP4hgIsNzZ8mtfFPayi8KlDWRQw3gj7ENRgxjbm4HxJgrdDNobuguc8EPQ3SccWXGTsZytLEeOHJXskiGlH4oEf"
实现异步请求
结构设计
底层Socket是天然支持异步的,因为发送和接收是可以同时进行的,不会互相影响。
要实现异步请求的效果,上层API只要做到以下几点:
- 请求发送后,不会阻塞当前执行线程
- 响应到达后可以触发回调
- 超时(指定时间内没有收到响应)也可以触发回调
实现方式
- 请求接口发送请求后返回Future对象,可选择同步等待
- 客户端保留请求和对应的callback
- 服务端响应的时候返回请求ID
- 客户端根据ID获取关联请求,执行callback。
首先,项目结构图如下:
1.划线部分是废弃类
2.【变更】解码方式修改,新增通用的MessageDecoder可供双方解码,其继承于ByteToMessageDecoder。
3.【新增】新增HelloResponse
4.【新增】新增通用MessageEncoder,继承于MessageToByteEncoder
代码实现
- MessageDecoder.java
相比前面直接实现的,这里不用去管理缓存。另外,这里解析好的消息会写入List,但它其实是逐个传给下一个Handler。
public class MessageDecoder extends ByteToMessageDecoder {
private static final int HEADER_LENGTH = 4; //消息头部长度
@Override
protected void decode(ChannelHandlerContext ctx, ByteBuf in, List<Object> out) throws Exception {
// 检查是否足够的字节来读取一个消息头
while (in.readableBytes() >= HEADER_LENGTH) {
in.markReaderIndex(); // 标记当前读取位置,便于重置
// 读取消息头部
byte[] headerBytes = new byte[4];
in.readBytes(headerBytes);
// 获取类型
int type = headerBytes[0] & 0xFF;
// 获取消息体长度
int bodyLength = ((headerBytes[1] & 0xFF) << 16) |
((headerBytes[2] & 0xFF) << 8) |
(headerBytes[3] & 0xFF);
// 检查缓冲区中的数据是否足够读取整个消息体
if (in.readableBytes() < bodyLength) {
in.resetReaderIndex(); // 重置读指针,等待更多数据
break;
}
// 读取消息体
byte[] body = new byte[bodyLength];
in.readBytes(body);
// 处理消息,根据消息头中的类型,解析成不同的对象
try {
Object msg = null;
if(type == 1) {
msg = HelloRequest.parseFrom(body);
} else if(type == 2) {
msg = HelloResponse.parseFrom(body);
} else {
System.out.println("未知消息:"+new String(body, CharsetUtil.UTF_8));
}
if(Objects.nonNull(msg)) {
out.add(msg);
}
} catch (Exception e) {
System.out.println("解析失败: " + new String(body, CharsetUtil.UTF_8));
}
}
}
}
2.MessageEncoder.java
ProtoBuf生成的类可以调用toByteArray()序列化成字节数组。这样消息体的二进制数据就有了。
而消息头则根据消息类型和消息体长度进行构建
public class MessageEncoder extends MessageToByteEncoder<Object> {
@Override
protected void encode(ChannelHandlerContext ctx, Object msg, ByteBuf out) throws Exception {
if(!(msg instanceof GeneratedMessage)) {
System.out.println("未知类型:"+msg.getClass());
return;
}
int type = 0;
if(msg instanceof HelloRequest) {
type = 1;
} else if(msg instanceof HelloResponse) {
type = 2;
}
byte[] bodyBytes = ((GeneratedMessage) msg).toByteArray();
int length = bodyBytes.length;
byte[] lengthBytes = new byte[3];
lengthBytes[0] = (byte) (length >> 16);
lengthBytes[1] = (byte) (length >> 8);
lengthBytes[2] = (byte) length;
out.writeByte(type);
out.writeBytes(lengthBytes);
out.writeBytes(bodyBytes);
}
}
3.ServerMessageHandler
服务器得到Decoder解析好的消息后,间隔一段时间(1-4秒)发回响应内容。
public class ServerMessageHandler extends SimpleChannelInboundHandler<Object> {
//异步线程
private ScheduledExecutorService mockRequestHandler = Executors.newSingleThreadScheduledExecutor();
@Override
protected void channelRead0(ChannelHandlerContext ctx, Object msg) throws Exception {
if(msg instanceof HelloRequest) {
System.out.println("收到消息:"+msg);
HelloRequest request = (HelloRequest) msg;
//使用处理线程,不阻塞NIO线程
//模拟处理请求,处理时间随机1~4秒
mockRequestHandler.schedule(() -> {
ctx.writeAndFlush(HelloResponse.newBuilder()
.setRequestId(request.getRequestId())
.setStatus(200)
.setData("Handled:"+request.getContent()) //增加一个前缀,表示服务器已处理
.build()
);
}, RandomUtil.randomInt(1, 4), TimeUnit.SECONDS);
}
}
}
4.EchoServer
服务端启动类,配置Handler,启动端口监听。
public class EchoServer {
private final int port;
public EchoServer(int port) {
this.port = port;
}
public void start() throws Exception {
EventLoopGroup bossGroup = new NioEventLoopGroup(); //处理CONNECT的线程
EventLoopGroup workerGroup = new NioEventLoopGroup(4); //Worker线程
try {
ServerBootstrap b = new ServerBootstrap();
b.group(bossGroup, workerGroup)
.channel(NioServerSocketChannel.class)
.localAddress(port)
.childHandler(new ChannelInitializer<SocketChannel>() {
protected void initChannel(SocketChannel socketChannel) throws Exception {
socketChannel.pipeline()
.addLast(new MessageDecoder())
.addLast(new MessageEncoder())
.addLast(new ServerMessageHandler());
}
});
ChannelFuture f = b.bind().sync(); //开始监听
System.out.println("启动监听:"+port);
f.channel().closeFuture().sync(); //阻塞直到程序退出
} finally {
bossGroup.shutdownGracefully().sync();
workerGroup.shutdownGracefully().sync();
}
}
public static void main(String[] args) throws Exception {
new EchoServer(9090).start();
}
}
5.ClientMessageHandler
对接ClientApi
1.连接建立后告知ClientApi
2.收到响应后提交给ClientApi
public class ClientMessageHandler extends SimpleChannelInboundHandler<Object> {
@Override
protected void channelRead0(ChannelHandlerContext ctx, Object msg) throws Exception {
if(msg instanceof HelloResponse) {
// System.out.println("收到消息:"+msg);
//收到响应内容,则触发回调
ClientApi clientApi = Container.getClientApi();
if(Objects.nonNull(clientApi)) {
clientApi.onResponse((HelloResponse) msg);
}
} else {
System.out.println("未知消息:"+msg);
}
}
@Override
public void channelActive(ChannelHandlerContext ctx) throws Exception {
super.channelActive(ctx);
//连接成功,告知ClientApi
ClientApi clientApi = Container.getClientApi();
if(Objects.nonNull(clientApi)) {
clientApi.onConnected(ctx);
}
}
}
6.EchoClient
客户端启动类,配置编解码类和消息处理类,最后连接到目标地址。
注意:这里没有main入口,入口在Test类。
public class EchoClient {
private final String host;
private final int port;
public EchoClient(String host, int port) {
this.host = host;
this.port = port; } public void start() throws Exception { EventLoopGroup group = new NioEventLoopGroup(); try { Bootstrap b = new Bootstrap(); b.group(group) .channel(NioSocketChannel.class) .remoteAddress(new InetSocketAddress(host, port)) .handler(new ChannelInitializer<SocketChannel>() { protected void initChannel(SocketChannel socketChannel) throws Exception { socketChannel.pipeline() .addLast(new MessageDecoder()) .addLast(new MessageEncoder()) .addLast(new ClientMessageHandler()); } }); ChannelFuture f = b.connect().sync(); System.out.println("开始连接"); f.channel().closeFuture().sync(); } finally { group.shutdownGracefully().sync(); } } }
7.Container
很简单,就是一个静态类。用来存放ClientApi的引用
public class Container {
public static ClientApi clientApi;
public static void setClientApi(ClientApi clientApi) {
Container.clientApi = clientApi;
}
public static ClientApi getClientApi() {
return clientApi;
}
}
8.ClientApi
ClientApi负责建立连接,发送请求,回调响应。支持同步和异步两种请求方式。
public class ClientApi {
private final String host;
private final int port;
private final Map<String, CompletableFuture<HelloResponse>> waitingRequests = new HashMap<>();
private final ScheduledExecutorService timer = Executors.newSingleThreadScheduledExecutor();
private final ReentrantLock lock = new ReentrantLock();
private ChannelHandlerContext ctx = null;
private CompletableFuture<ChannelHandlerContext> waitConnectionFuture;
ClientApi(String host, int port) {
this.host = host;
this.port = port;
//初始化后注册到Container中,方便其他类引用
Container.setClientApi(this);
}
//连接建立后,回调Context
public void onConnected(ChannelHandlerContext ctx) {
lock.lock();
try {
if(waitConnectionFuture != null) {
waitConnectionFuture.complete(ctx);
waitConnectionFuture = null;
}
} finally {
lock.unlock();
}
}
//获取连接
public ChannelHandlerContext getConnection() throws Exception {
lock.lock();
try {
//连接已存在,直接发
if(ctx != null) {
return ctx;
}
//连接不存在,建立连接
waitConnectionFuture = new CompletableFuture<>();
new Thread(()-> {
try {
new EchoClient(host, port).start(); //这个会阻塞当前线程,所以另启线程
} catch (Exception e) {
e.printStackTrace();
//连接断开,也触发回调
if(waitConnectionFuture != null) {
waitConnectionFuture.completeExceptionally(e);
}
}
}).start();
} finally {
lock.unlock();
}
//get()等待之前,需要释放锁
ctx = waitConnectionFuture.get();
return ctx;
}
public CompletableFuture<HelloResponse> baseRequest(HelloRequest request) {
//先注册回调
CompletableFuture<HelloResponse> future = new CompletableFuture<>();
addToMap(request.getRequestId(), future);
//再发送请求
try {
getConnection().writeAndFlush(request);
System.out.println("发出消息:"+request);
} catch (Exception e) {
removeFromMap(request.getRequestId());
throw new RuntimeException("请求错误:"+e);
}
//添加超时,防止服务器没响应,造成泄露
timer.schedule(() -> timeout(request.getRequestId()), 5, TimeUnit.SECONDS);
return future;
}
//同步请求
public HelloResponse sendRequest(HelloRequest request) throws Exception {
return this.baseRequest(request).get();
}
//异步请求
public void sendRequestAsync(HelloRequest request, Function<HelloResponse, Boolean> callback) {
this.baseRequest(request).thenApply(callback);
}
private void addToMap(String requestId, CompletableFuture<HelloResponse> future) {
lock.lock();
try {
waitingRequests.put(requestId, future);
} finally {
lock.unlock();
}
}
private void removeFromMap(String requestId) {
lock.lock();
try {
waitingRequests.remove(requestId);
} finally {
lock.unlock();
}
}
public void timeout(String requestId) {
lock.lock();
try {
CompletableFuture<HelloResponse> future1 = waitingRequests.get(requestId);
if(Objects.nonNull(future1)) {
future1.completeExceptionally(new RuntimeException("请求超时"));
}
} finally {
lock.unlock();
}
}
public void onResponse(HelloResponse response) {
lock.lock();
try {
//收到响应后,根据请求ID获取回调。
CompletableFuture<HelloResponse> future1 = waitingRequests.get(response.getRequestId());
if(Objects.nonNull(future1)) {
future1.complete(response);
}
} finally {
lock.unlock();
}
}
public void close() {
if(ctx != null) {
ctx.close();
}
}
}
9.测试类
测试类,使用ClientApi,发送请求。
下面代码分别是同步发送5个请求和异步发送5个请求
public class Test {
public static void main(String[] args) {
ClientApi clientApi = new ClientApi("127.0.0.1", 9090);
for (int i = 0; i < 5; i++) {
try {
HelloRequest request = buildHelloRequest();
//同步请求,收到响应后才会发下一个请求
HelloResponse response = clientApi.sendRequest(request);
System.out.println("同步收到:" + response);
//异步请求,发送完成即可发送下一个请求
// clientApi.sendRequestAsync(request, response2 -> {
// System.out.println("异步收到:"+response2);
// return true;
// });
} catch (Exception e) {
e.printStackTrace();
}
}
}
//构建请求
public static HelloRequest buildHelloRequest() {
int randomStrLength = RandomUtil.randomInt(100, 200);
String msgId = UUID.randomUUID().toString();
String content = RandomUtil.randomString(randomStrLength);
HelloRequest request = HelloRequest.newBuilder()
.setRequestId(msgId)
.setContent(content)
.build();
return request;
}
}
运行效果
1)同步请求
可以看到,只有收到前一个请求的响应后,才会发下一个请求
开始连接
发出消息:requestId: "7ac9008f-8532-4740-bb1f-f5fe2d60fd62"
content: "zYU1oWKiJdLX2K87J0306Xdeq0BmlkijpD1p6t831A3b2fEP0JCZT9QqaT7oWnCpoKMIpZRZ2gJWbSvEIhxs56m8Zr0YhsvaPdQj1x8f4Q1HuLh"
同步收到:requestId: "7ac9008f-8532-4740-bb1f-f5fe2d60fd62"
status: 200
data: "Handled:zYU1oWKiJdLX2K87J0306Xdeq0BmlkijpD1p6t831A3b2fEP0JCZT9QqaT7oWnCpoKMIpZRZ2gJWbSvEIhxs56m8Zr0YhsvaPdQj1x8f4Q1HuLh"
发出消息:requestId: "d74a8a83-28dd-4e40-9956-dfdd3d890bda"
content: "83LzAdgxQ8MYz1CmzRXfEn3ibz9WqiJHcQCRtkE4dCEZUnW44UNGfKtHR0nBNE7al7PdvdexEDDTth3Aoy6mati8TVOP54xRUT26MAaV0DP0UhU7V7QWaMfiKwV2oVpxMPqg2thwNdd5WCG53"
同步收到:requestId: "d74a8a83-28dd-4e40-9956-dfdd3d890bda"
status: 200
data: "Handled:83LzAdgxQ8MYz1CmzRXfEn3ibz9WqiJHcQCRtkE4dCEZUnW44UNGfKtHR0nBNE7al7PdvdexEDDTth3Aoy6mati8TVOP54xRUT26MAaV0DP0UhU7V7QWaMfiKwV2oVpxMPqg2thwNdd5WCG53"
发出消息:requestId: "e47f4135-dc92-4a25-9fed-ca7b4ced41f5"
content: "1St23ktz7nhXcICb0Yqo3QoLNRoeKJ4V4jKcaO8psUZRXnhXtIUG2WZC7d0TCbS221pUxqTlTDojSbtQvCzjIcL3JTgPkBg46rk8uRIpX3yvx0RHmZhwjBkpPoCTtWfTzk6r5SK0SK7g0QAjQacULIDXS5K1Z1U9q"
同步收到:requestId: "e47f4135-dc92-4a25-9fed-ca7b4ced41f5"
status: 200
data: "Handled:1St23ktz7nhXcICb0Yqo3QoLNRoeKJ4V4jKcaO8psUZRXnhXtIUG2WZC7d0TCbS221pUxqTlTDojSbtQvCzjIcL3JTgPkBg46rk8uRIpX3yvx0RHmZhwjBkpPoCTtWfTzk6r5SK0SK7g0QAjQacULIDXS5K1Z1U9q"
发出消息:requestId: "5122e929-be13-488e-b3ea-6e5acf7ebbbc"
content: "pj9sWOqhFGzplbUNieLOOzMKzSUEumgd2rMzR1cO4GwrmcXHb5vma32LbuHVQl8tkigKMHk9HCKM9xnUAdbCqopeTzbo0ixQkGzclud78hVFTV4PM2qYZDeWMBRDMrUXOJS0sCIxgyGFudz7XUGfJNSuJjio8dch8JPDRmHkZsABRxobZeafxiqGT"
同步收到:requestId: "5122e929-be13-488e-b3ea-6e5acf7ebbbc"
status: 200
data: "Handled:pj9sWOqhFGzplbUNieLOOzMKzSUEumgd2rMzR1cO4GwrmcXHb5vma32LbuHVQl8tkigKMHk9HCKM9xnUAdbCqopeTzbo0ixQkGzclud78hVFTV4PM2qYZDeWMBRDMrUXOJS0sCIxgyGFudz7XUGfJNSuJjio8dch8JPDRmHkZsABRxobZeafxiqGT"
发出消息:requestId: "0cd23413-303d-4414-8cf1-20bd46a691d2"
content: "YWnQxVh0Z4yLPQeM6q3aiz7JYD6fEqZHFiE45KgebiZlwW7DlYnhZTZ7sG4rZqrvsHXQ65PCoN569kfJMHuJFp9kqnlBKeJ1iawYBFQfI5EqspxsaB7vkMuC1vA5ula2jwagoQoU6Yk0gi0EKEX1fpLIYvtYdMqTWjAfLFqc5s8yjPr0G"
同步收到:requestId: "0cd23413-303d-4414-8cf1-20bd46a691d2"
status: 200
data: "Handled:YWnQxVh0Z4yLPQeM6q3aiz7JYD6fEqZHFiE45KgebiZlwW7DlYnhZTZ7sG4rZqrvsHXQ65PCoN569kfJMHuJFp9kqnlBKeJ1iawYBFQfI5EqspxsaB7vkMuC1vA5ula2jwagoQoU6Yk0gi0EKEX1fpLIYvtYdMqTWjAfLFqc5s8yjPr0G"
2)异步请求
可以看到5个请求会直接发出,不会等待响应。响应顺序也跟请求顺序不一样。
开始连接
发出消息:requestId: "2096b54f-825c-4fdf-817e-97b3a4b99fb2"
content: "ckp1dXcnYItdXafrURU6gJ9b5qW19rqPKLB22qlA2sHRfkSZEpmT4qi0TOAaDbM43v62svI1K6IccnlJtCjlpcu8RAdpfuO5hNBWsXpOaSGUgY4loLNlNFIDE5o7juhfCD2skV2"
发出消息:requestId: "e8c501e0-a4e2-4972-a6be-f92539141252"
content: "ZpOz1YK6e9VOrX69xqNovUXfapY6Ito7z6LlsM6o1Vzeo1hibzvOcxAzYD8hIsOFvGAqk024XbL7yidlgPk4F9GId6ydRxzjjNdg8csxG9FdBXzzr6xuESJ"
发出消息:requestId: "2c56fc02-5fd8-4c17-bca5-ab0def66493c"
content: "FzoUaLvHxA0Tm7eU4GL9bIE6mEMNRIUSZILPLiREPXGfhcgoasYd1W5jEfAooE697LQr2DMw6fBdwEqHunQcl6doxrnxSQAZorHztHvyKXAFmbnF3aDkYgO82HaHGXuC"
发出消息:requestId: "477cbb82-77c0-4270-b161-555da7b6a5e2"
content: "Ow81y5qPraVcobOiZ6sCH72jJGNC0784ox5crQYP5fZ6CXoWphRdC9WW4NKSiChbci6aGutnWJbO1HlpR0FV4m9qahbWGkFI0Zr2uvMbuaj8SPpH6X"
发出消息:requestId: "98db5068-1a85-455f-8fea-4b6c8a562776"
content: "nWnNIaJFt1otie04SWoaoN08f2BOuTMyRbtFuEhj0LiYilRjeKswzqrbKlze30ZBFNIuvEz6P97rP9lM5bkuDYLv1QuKOd1wctfeF9K2RbKh6hvOfgHE5wl2xUk0B6nBFK5fI1sdj3hhoiPLApQZjGzFaSHZGVtLdM4yPBC6BhmsNCPkAo2AxcQ0iZuVEHkihs"
异步收到:requestId: "2096b54f-825c-4fdf-817e-97b3a4b99fb2"
status: 200
data: "Handled:ckp1dXcnYItdXafrURU6gJ9b5qW19rqPKLB22qlA2sHRfkSZEpmT4qi0TOAaDbM43v62svI1K6IccnlJtCjlpcu8RAdpfuO5hNBWsXpOaSGUgY4loLNlNFIDE5o7juhfCD2skV2"
异步收到:requestId: "e8c501e0-a4e2-4972-a6be-f92539141252"
status: 200
data: "Handled:ZpOz1YK6e9VOrX69xqNovUXfapY6Ito7z6LlsM6o1Vzeo1hibzvOcxAzYD8hIsOFvGAqk024XbL7yidlgPk4F9GId6ydRxzjjNdg8csxG9FdBXzzr6xuESJ"
异步收到:requestId: "477cbb82-77c0-4270-b161-555da7b6a5e2"
status: 200
data: "Handled:Ow81y5qPraVcobOiZ6sCH72jJGNC0784ox5crQYP5fZ6CXoWphRdC9WW4NKSiChbci6aGutnWJbO1HlpR0FV4m9qahbWGkFI0Zr2uvMbuaj8SPpH6X"
异步收到:requestId: "2c56fc02-5fd8-4c17-bca5-ab0def66493c"
status: 200
data: "Handled:FzoUaLvHxA0Tm7eU4GL9bIE6mEMNRIUSZILPLiREPXGfhcgoasYd1W5jEfAooE697LQr2DMw6fBdwEqHunQcl6doxrnxSQAZorHztHvyKXAFmbnF3aDkYgO82HaHGXuC"
异步收到:requestId: "98db5068-1a85-455f-8fea-4b6c8a562776"
status: 200
data: "Handled:nWnNIaJFt1otie04SWoaoN08f2BOuTMyRbtFuEhj0LiYilRjeKswzqrbKlze30ZBFNIuvEz6P97rP9lM5bkuDYLv1QuKOd1wctfeF9K2RbKh6hvOfgHE5wl2xUk0B6nBFK5fI1sdj3hhoiPLApQZjGzFaSHZGVtLdM4yPBC6BhmsNCPkAo2AxcQ0iZuVEHkihs"
实现订阅发布
//TBD